diffusers/pr_14407/zh/_app/immutable/nodes/49.DT72jZHs.js

import"../chunks/DsnmJJEf.js";import{i as x,h as B,C as V,H as n,a,b as o,E as k,s as F}from"../chunks/DdZvggmf.js";import{p as Y,o as N,s as l,f as E,a as v,b as z,c as R,n as C}from"../chunks/BbekZcyp.js";import{H as i}from"../chunks/BcnRgdDK.js";const S='{"title":"文本反转（Textual Inversion）","local":"文本反转textual-inversion","sections":[{"title":"脚本参数","local":"脚本参数","sections":[],"depth":2},{"title":"训练脚本","local":"训练脚本","sections":[],"depth":2},{"title":"启动脚本","local":"启动脚本","sections":[],"depth":2},{"title":"后续步骤","local":"后续步骤","sections":[],"depth":2}],"depth":1}';var Q=R('<meta name="hf:doc:metadata"/>'),A=R('<p></p> <!> <!> <p><a href="https://hf.co/papers/2208.01618" rel="nofollow">文本反转</a>是一种训练技术，仅需少量示例图像即可个性化图像生成模型。该技术通过学习和更新文本嵌入（新嵌入会绑定到提示中必须使用的特殊词汇）来匹配您提供的示例图像。</p> <p>如果在显存有限的GPU上训练，建议在训练命令中启用<code>gradient_checkpointing</code>和<code>mixed_precision</code>参数。您还可以通过<a href="../optimization/xformers">xFormers</a>使用内存高效注意力机制来减少内存占用。</p> <p>本指南将探索<a href="https://github.com/huggingface/diffusers/blob/main/examples/textual_inversion/textual_inversion.py" rel="nofollow">textual_inversion.py</a>脚本，帮助您更熟悉其工作原理，并了解如何根据自身需求进行调整。</p> <p>运行脚本前，请确保从源码安装库：</p> <!> <p>进入包含训练脚本的示例目录，并安装所需依赖：</p> <!> <blockquote class="tip"><p>🤗 Accelerate 是一个帮助您在多GPU/TPU或混合精度环境下训练的工具库。它会根据硬件和环境自动配置训练设置。查看🤗 Accelerate <a href="https://huggingface.co/docs/accelerate/quicktour" rel="nofollow">快速入门</a>了解更多。</p></blockquote> <p>初始化🤗 Accelerate环境：</p> <!> <p>要设置默认的🤗 Accelerate环境（不选择任何配置）：</p> <!> <p>如果您的环境不支持交互式shell（如notebook），可以使用：</p> <!> <p>最后，如果想在自定义数据集上训练模型，请参阅<a href="create_dataset">创建训练数据集</a>指南，了解如何创建适用于训练脚本的数据集。</p> <blockquote class="tip"><p>以下部分重点介绍训练脚本中需要理解的关键修改点，但未涵盖脚本所有细节。如需深入了解，可随时查阅<a href="https://github.com/huggingface/diffusers/blob/main/examples/textual_inversion/textual_inversion.py" rel="nofollow">脚本源码</a>，如有疑问欢迎反馈。</p></blockquote> <!> <p>训练脚本包含众多参数，便于您定制训练过程。所有参数及其说明都列在<a href="https://github.com/huggingface/diffusers/blob/839c2a5ece0af4e75530cb520d77bc7ed8acf474/examples/textual_inversion/textual_inversion.py#L176" rel="nofollow"><code>parse_args()</code></a>函数中。Diffusers为每个参数提供了默认值（如训练批次大小和学习率），但您可以通过训练命令自由调整这些值。</p> <p>例如，将梯度累积步数增加到默认值1以上：</p> <!> <p>其他需要指定的基础重要参数包括：</p> <ul><li><code>--pretrained_model_name_or_path</code>：Hub上的模型名称或本地预训练模型路径</li> <li><code>--train_data_dir</code>：包含训练数据集（示例图像）的文件夹路径</li> <li><code>--output_dir</code>：训练模型保存位置</li> <li><code>--push_to_hub</code>：是否将训练好的模型推送至Hub</li> <li><code>--checkpointing_steps</code>：训练过程中保存检查点的频率；若训练意外中断，可通过在命令中添加<code>--resume_from_checkpoint</code>从该检查点恢复训练</li> <li><code>--num_vectors</code>：学习嵌入的向量数量；增加此参数可提升模型效果，但会提高训练成本</li> <li><code>--placeholder_token</code>：绑定学习嵌入的特殊词汇（推理时需在提示中使用该词）</li> <li><code>--initializer_token</code>：大致描述训练目标的单字词汇（如物体或风格）</li> <li><code>--learnable_property</code>：训练目标是学习新”风格”（如梵高画风）还是”物体”（如您的宠物狗）</li></ul> <!> <p>与其他训练脚本不同，textual_inversion.py包含自定义数据集类<a href="https://github.com/huggingface/diffusers/blob/b81c69e489aad3a0ba73798c459a33990dc4379c/examples/textual_inversion/textual_inversion.py#L487" rel="nofollow"><code>TextualInversionDataset</code></a>，用于创建数据集。您可以自定义图像尺寸、占位符词汇、插值方法、是否裁剪图像等。如需修改数据集创建方式，可调整<code>TextualInversionDataset</code>类。</p> <p>接下来，在<a href="https://github.com/huggingface/diffusers/blob/839c2a5ece0af4e75530cb520d77bc7ed8acf474/examples/textual_inversion/textual_inversion.py#L573" rel="nofollow"><code>main()</code></a>函数中可找到数据集预处理代码和训练循环。</p> <p>脚本首先加载<a href="https://github.com/huggingface/diffusers/blob/b81c69e489aad3a0ba73798c459a33990dc4379c/examples/textual_inversion/textual_inversion.py#L616" rel="nofollow">tokenizer</a>、<a href="https://github.com/huggingface/diffusers/blob/b81c69e489aad3a0ba73798c459a33990dc4379c/examples/textual_inversion/textual_inversion.py#L622" rel="nofollow">scheduler和模型</a>：</p> <!> <p>随后将特殊<a href="https://github.com/huggingface/diffusers/blob/b81c69e489aad3a0ba73798c459a33990dc4379c/examples/textual_inversion/textual_inversion.py#L632" rel="nofollow">占位符词汇</a>加入tokenizer，并调整嵌入层以适配新词汇。</p> <p>接着，脚本通过<code>TextualInversionDataset</code><a href="https://github.com/huggingface/diffusers/blob/b81c69e489aad3a0ba73798c459a33990dc4379c/examples/textual_inversion/textual_inversion.py#L716" rel="nofollow">创建数据集</a>：</p> <!> <p>最后，<a href="https://github.com/huggingface/diffusers/blob/b81c69e489aad3a0ba73798c459a33990dc4379c/examples/textual_inversion/textual_inversion.py#L784" rel="nofollow">训练循环</a>处理从预测噪声残差到更新特殊占位符词汇嵌入权重的所有流程。</p> <p>如需深入了解训练循环工作原理，请参阅<a href="../using-diffusers/write_own_pipeline">理解管道、模型与调度器</a>教程，该教程解析了去噪过程的基本模式。</p> <!> <p>完成所有修改或确认默认配置后，即可启动训练脚本！🚀</p> <p>本指南将下载<a href="https://huggingface.co/datasets/diffusers/cat_toy_example" rel="nofollow">猫玩具</a>的示例图像并存储在目录中。当然，您也可以创建和使用自己的数据集（参见<a href="create_dataset">创建训练数据集</a>指南）。</p> <!> <p>设置环境变量<code>MODEL_NAME</code>为Hub上的模型ID或本地模型路径，<code>DATA_DIR</code>为刚下载的猫图像路径。脚本会将以下文件保存至您的仓库：</p> <ul><li><code>learned_embeds.bin</code>：与示例图像对应的学习嵌入向量</li> <li><code>token_identifier.txt</code>：特殊占位符词汇</li> <li><code>type_of_concept.txt</code>：训练概念类型（“object”或”style”）</li></ul> <blockquote class="warning"><p>在单块V100 GPU上完整训练约需1小时。</p></blockquote> <p>启动脚本前还有最后一步。如果想实时观察训练过程，可以定期保存生成图像。在训练命令中添加以下参数：</p> <!> <!> <p>训练完成后，可以像这样使用新模型进行推理：</p> <!> <!> <p>恭喜您成功训练了自己的文本反转模型！🎉 如需了解更多使用技巧，以下指南可能会有所帮助：</p> <ul><li>学习如何<a href="../using-diffusers/loading_adapters">加载文本反转嵌入</a>，并将其用作负面嵌入</li> <li>学习如何将<a href="textual_inversion_inference">文本反转</a>应用于Stable Diffusion 1/2和Stable Diffusion XL的推理</li></ul> <!> <p></p>',1);function $(W,G){Y(G,!1),N(()=>{new URLSearchParams(window.location.search).get("fw")}),x();var c=A();B("11u6mld",e=>{var s=Q();F(s,"content",S),v(e,s)});var r=l(E(c),2);V(r,{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"});var p=l(r,2);n(p,{title:"文本反转（Textual Inversion）",local:"文本反转textual-inversion",headingTag:"h1"});var d=l(p,10);a(d,{code:"Z2l0JTIwY2xvbmUlMjBodHRwcyUzQSUyRiUyRmdpdGh1Yi5jb20lMkZodWdnaW5nZmFjZSUyRmRpZmZ1c2VycyUwQWNkJTIwZGlmZnVzZXJzJTBBcGlwJTIwaW5zdGFsbCUyMC4=",highlighted:`git <span class="hljs-built_in">clone</span> https://github.com/huggingface/diffusers
<span class="hljs-built_in">cd</span> diffusers
pip install .`,lang:"bash",wrap:!1});var h=l(d,4);o(h,{id:"installation",options:["PyTorch"],children:(e,s)=>{i(e,{id:"installation",option:"PyTorch",children:(t,X)=>{a(t,{code:"Y2QlMjBleGFtcGxlcyUyRnRleHR1YWxfaW52ZXJzaW9uJTBBcGlwJTIwaW5zdGFsbCUyMC1yJTIwcmVxdWlyZW1lbnRzLnR4dA==",highlighted:`<span class="hljs-built_in">cd</span> examples/textual_inversion
pip install -r requirements.txt`,lang:"bash",wrap:!1})},$$slots:{default:!0}})},$$slots:{default:!0}});var u=l(h,6);a(u,{code:"YWNjZWxlcmF0ZSUyMGNvbmZpZw==",highlighted:"accelerate config",lang:"bash",wrap:!1});var M=l(u,4);a(M,{code:"YWNjZWxlcmF0ZSUyMGNvbmZpZyUyMGRlZmF1bHQ=",highlighted:"accelerate config default",lang:"bash",wrap:!1});var y=l(M,4);a(y,{code:"ZnJvbSUyMGFjY2VsZXJhdGUudXRpbHMlMjBpbXBvcnQlMjB3cml0ZV9iYXNpY19jb25maWclMEElMEF3cml0ZV9iYXNpY19jb25maWcoKQ==",highlighted:`<span class="hljs-keyword">from</span> accelerate.utils <span class="hljs-keyword">import</span> write_basic_config

write_basic_config()`,lang:"py",wrap:!1});var b=l(y,6);n(b,{title:"脚本参数",local:"脚本参数",headingTag:"h2"});var J=l(b,6);a(J,{code:"YWNjZWxlcmF0ZSUyMGxhdW5jaCUyMHRleHR1YWxfaW52ZXJzaW9uLnB5JTIwJTVDJTBBJTIwJTIwLS1ncmFkaWVudF9hY2N1bXVsYXRpb25fc3RlcHMlM0Q0",highlighted:`accelerate launch textual_inversion.py \\
  --gradient_accumulation_steps=4`,lang:"bash",wrap:!1});var m=l(J,6);n(m,{title:"训练脚本",local:"训练脚本",headingTag:"h2"});var f=l(m,8);a(f,{code:"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",highlighted:`<span class="hljs-comment"># 加载tokenizer</span>
<span class="hljs-keyword">if</span> args.tokenizer_name:
    tokenizer = CLIPTokenizer.from_pretrained(args.tokenizer_name)
<span class="hljs-keyword">elif</span> args.pretrained_model_name_or_path:
    tokenizer = CLIPTokenizer.from_pretrained(args.pretrained_model_name_or_path, subfolder=<span class="hljs-string">&quot;tokenizer&quot;</span>)

<span class="hljs-comment"># 加载scheduler和模型</span>
noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder=<span class="hljs-string">&quot;scheduler&quot;</span>)
text_encoder = CLIPTextModel.from_pretrained(
    args.pretrained_model_name_or_path, subfolder=<span class="hljs-string">&quot;text_encoder&quot;</span>, revision=args.revision
)
vae = AutoencoderKL.from_pretrained(args.pretrained_model_name_or_path, subfolder=<span class="hljs-string">&quot;vae&quot;</span>, revision=args.revision)
unet = UNet2DConditionModel.from_pretrained(
    args.pretrained_model_name_or_path, subfolder=<span class="hljs-string">&quot;unet&quot;</span>, revision=args.revision
)`,lang:"py",wrap:!1});var _=l(f,6);a(_,{code:"dHJhaW5fZGF0YXNldCUyMCUzRCUyMFRleHR1YWxJbnZlcnNpb25EYXRhc2V0KCUwQSUyMCUyMCUyMCUyMGRhdGFfcm9vdCUzRGFyZ3MudHJhaW5fZGF0YV9kaXIlMkMlMEElMjAlMjAlMjAlMjB0b2tlbml6ZXIlM0R0b2tlbml6ZXIlMkMlMEElMjAlMjAlMjAlMjBzaXplJTNEYXJncy5yZXNvbHV0aW9uJTJDJTBBJTIwJTIwJTIwJTIwcGxhY2Vob2xkZXJfdG9rZW4lM0QoJTIyJTIwJTIyLmpvaW4odG9rZW5pemVyLmNvbnZlcnRfaWRzX3RvX3Rva2VucyhwbGFjZWhvbGRlcl90b2tlbl9pZHMpKSklMkMlMEElMjAlMjAlMjAlMjByZXBlYXRzJTNEYXJncy5yZXBlYXRzJTJDJTBBJTIwJTIwJTIwJTIwbGVhcm5hYmxlX3Byb3BlcnR5JTNEYXJncy5sZWFybmFibGVfcHJvcGVydHklMkMlMEElMjAlMjAlMjAlMjBjZW50ZXJfY3JvcCUzRGFyZ3MuY2VudGVyX2Nyb3AlMkMlMEElMjAlMjAlMjAlMjBzZXQlM0QlMjJ0cmFpbiUyMiUyQyUwQSklMEF0cmFpbl9kYXRhbG9hZGVyJTIwJTNEJTIwdG9yY2gudXRpbHMuZGF0YS5EYXRhTG9hZGVyKCUwQSUyMCUyMCUyMCUyMHRyYWluX2RhdGFzZXQlMkMlMjBiYXRjaF9zaXplJTNEYXJncy50cmFpbl9iYXRjaF9zaXplJTJDJTIwc2h1ZmZsZSUzRFRydWUlMkMlMjBudW1fd29ya2VycyUzRGFyZ3MuZGF0YWxvYWRlcl9udW1fd29ya2VycyUwQSk=",highlighted:`train_dataset = TextualInversionDataset(
    data_root=args.train_data_dir,
    tokenizer=tokenizer,
    size=args.resolution,
    placeholder_token=(<span class="hljs-string">&quot; &quot;</span>.join(tokenizer.convert_ids_to_tokens(placeholder_token_ids))),
    repeats=args.repeats,
    learnable_property=args.learnable_property,
    center_crop=args.center_crop,
    <span class="hljs-built_in">set</span>=<span class="hljs-string">&quot;train&quot;</span>,
)
train_dataloader = torch.utils.data.DataLoader(
    train_dataset, batch_size=args.train_batch_size, shuffle=<span class="hljs-literal">True</span>, num_workers=args.dataloader_num_workers
)`,lang:"py",wrap:!1});var g=l(_,6);n(g,{title:"启动脚本",local:"启动脚本",headingTag:"h2"});var U=l(g,6);a(U,{code:"ZnJvbSUyMGh1Z2dpbmdmYWNlX2h1YiUyMGltcG9ydCUyMHNuYXBzaG90X2Rvd25sb2FkJTBBJTBBbG9jYWxfZGlyJTIwJTNEJTIwJTIyLiUyRmNhdCUyMiUwQXNuYXBzaG90X2Rvd25sb2FkKCUwQSUyMCUyMCUyMCUyMCUyMmRpZmZ1c2VycyUyRmNhdF90b3lfZXhhbXBsZSUyMiUyQyUyMGxvY2FsX2RpciUzRGxvY2FsX2RpciUyQyUyMHJlcG9fdHlwZSUzRCUyMmRhdGFzZXQlMjIlMkMlMjBpZ25vcmVfcGF0dGVybnMlM0QlMjIuZ2l0YXR0cmlidXRlcyUyMiUwQSk=",highlighted:`<span class="hljs-keyword">from</span> huggingface_hub <span class="hljs-keyword">import</span> snapshot_download

local_dir = <span class="hljs-string">&quot;./cat&quot;</span>
snapshot_download(
    <span class="hljs-string">&quot;diffusers/cat_toy_example&quot;</span>, local_dir=local_dir, repo_type=<span class="hljs-string">&quot;dataset&quot;</span>, ignore_patterns=<span class="hljs-string">&quot;.gitattributes&quot;</span>
)`,lang:"py",wrap:!1});var T=l(U,10);a(T,{code:"LS12YWxpZGF0aW9uX3Byb21wdCUzRCUyMkElMjAlM0NjYXQtdG95JTNFJTIwdHJhaW4lMjIlMEEtLW51bV92YWxpZGF0aW9uX2ltYWdlcyUzRDQlMEEtLXZhbGlkYXRpb25fc3RlcHMlM0QxMDA=",highlighted:`--validation_prompt=<span class="hljs-string">&quot;A &lt;cat-toy&gt; train&quot;</span>
--num_validation_images=4
--validation_steps=100`,lang:"bash",wrap:!1});var Z=l(T,2);o(Z,{id:"training-inference",options:["PyTorch"],children:(e,s)=>{i(e,{id:"training-inference",option:"PyTorch",children:(t,X)=>{a(t,{code:"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",highlighted:`<span class="hljs-built_in">export</span> MODEL_NAME=<span class="hljs-string">&quot;stable-diffusion-v1-5/stable-diffusion-v1-5&quot;</span>
<span class="hljs-built_in">export</span> DATA_DIR=<span class="hljs-string">&quot;./cat&quot;</span>

accelerate launch textual_inversion.py \\
  --pretrained_model_name_or_path=<span class="hljs-variable">$MODEL_NAME</span> \\
  --train_data_dir=<span class="hljs-variable">$DATA_DIR</span> \\
  --learnable_property=<span class="hljs-string">&quot;object&quot;</span> \\
  --placeholder_token=<span class="hljs-string">&quot;&lt;cat-toy&gt;&quot;</span> \\
  --initializer_token=<span class="hljs-string">&quot;toy&quot;</span> \\
  --resolution=512 \\
  --train_batch_size=1 \\
  --gradient_accumulation_steps=4 \\
  --max_train_steps=3000 \\
  --learning_rate=5.0e-04 \\
  --scale_lr \\
  --lr_scheduler=<span class="hljs-string">&quot;constant&quot;</span> \\
  --lr_warmup_steps=0 \\
  --output_dir=<span class="hljs-string">&quot;textual_inversion_cat&quot;</span> \\
  --push_to_hub`,lang:"bash",wrap:!1})},$$slots:{default:!0}})},$$slots:{default:!0}});var w=l(Z,4);o(w,{id:"training-inference",options:["PyTorch"],children:(e,s)=>{i(e,{id:"training-inference",option:"PyTorch",children:(t,X)=>{a(t,{code:"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",highlighted:`<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> StableDiffusionPipeline
<span class="hljs-keyword">import</span> torch

pipeline = StableDiffusionPipeline.from_pretrained(<span class="hljs-string">&quot;stable-diffusion-v1-5/stable-diffusion-v1-5&quot;</span>, dtype=torch.float16).to(<span class="hljs-string">&quot;cuda&quot;</span>)
pipeline.load_textual_inversion(<span class="hljs-string">&quot;sd-concepts-library/cat-toy&quot;</span>)
image = pipeline(<span class="hljs-string">&quot;A &lt;cat-toy&gt; train&quot;</span>, num_inference_steps=<span class="hljs-number">50</span>).images[<span class="hljs-number">0</span>]
image.save(<span class="hljs-string">&quot;cat-train.png&quot;</span>)`,lang:"py",wrap:!1})},$$slots:{default:!0}})},$$slots:{default:!0}});var j=l(w,2);n(j,{title:"后续步骤",local:"后续步骤",headingTag:"h2"});var I=l(j,6);k(I,{source:"https://github.com/huggingface/diffusers/blob/main/docs/source/zh/training/text_inversion.md"}),C(2),v(W,c),z()}export{$ as component};